A combination of singular value decomposition and multivariate feature selection method for diagnosis of schizophrenia using fMRI. (May 2016)

• Record Type: Journal Article
• Authors: Juneja, Akanksha; Rana, Bharti; Agrawal, R.K.
• Is Part Of: Biomedical signal processing and control. Volume 27(2016)
• Page Start: 122
• Page End: 133
• Publication Date: 2016-05
• Subjects: Schizophrenia -- Computer-aided diagnosis -- Functional magnetic resonance imaging -- Singular value decomposition -- Multivariate forward feature selection; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2016.02.009
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Proposed a three-phase dimensionality reduction for CAD of schizophrenia using fMRI. Used spatial clustering on each 3-D spatial map created using GLM/ICA. Obtained discriminative features using SVD and a novel multivariate feature selection. Better performance than the existing methods is demonstrated on two datasets. Identified discriminative brain regions are consistent for both the datasets. Abstract: Schizophrenia is a severe psychiatric disorder which lacks any established diagnostic test and is currently diagnosed on the basis of externally observed behavioral symptoms. Functional magnetic resonance imaging (fMRI) is helpful in capturing abnormalities in brain activation patterns of schizophrenia patients in comparison to healthy subjects. Since the dimension of fMRI data is huge, while the number of samples is limited, dimensionality reduction is essential. Thus, this research work aims to utilize pattern recognition techniques to reduce the dimension of fMRI data for developing an effective computer-aided diagnosis method for schizophrenia. A three-phase method is proposed which involves spatial clustering of whole-brain voxels of individual 3-D spatial maps (β-maps or independent component score-maps), representation of each cluster using singular value decomposition followed by a novel hybrid multivariate forward feature selection method to obtain an optimal subset of relevant and non-redundant features for classification. A decision model is built using support vector machine classifier with leave-one-out cross-validation scheme. The measures, namely, sensitivity, specificity and classification accuracy are utilized to evaluate the performance of the decision model. The efficacy of the proposed method is evaluated on two distinct balanced datasets D1 and D2 (captured from 1.5 T and 3 T MRI scanners, respectively). D1 and D2 comprise of auditory oddball task fMRI data of schizophrenia patients and well age-matched healthy subjects, derived from publicly available FBIRN multisite dataset. Best classification accuracy of 92.6% and 94% are achieved for D1 and D2, respectively with the proposed method. The proposed method exhibits superior performance over the existing methods. In addition, discriminative brain regions, corresponding to the optimal subset of features, are identified and are in accordance with the literature. The proposed method is able to effectively classify schizophrenia patients and healthy subjects and thus, may be utilized as a diagnostic tool.